A Learning-Based Dependency to Constituency Conversion Algorithm for the Turkish Language

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5054–5062 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5054 A Learning-Based Dependency to Constituency Conversion Algorithm for the Turkish Language B¨us¸ra Mars¸an♥, O˘guz Kerem Yıldız♠, Aslı Kuzgun♥, Neslihan Cesur♥, Arife Bet¨ul Yenice♥ Ezgi Sanıyar♥, O˘guzhan Kuyrukc¸u♥, Bilge Nas Arıcan♥, Olcay Taner Yıldız♦ Starlang Yazılım Danıs¸manlık♥, Ahmet Keles¸o˘glu High School♠, Ozyegin University♦ Istanbul, Turkey {busra, oguz, asli, neslihan, arife, ezgi, oguzhan, bilge}@starlangyazilim.com, olcay.yildiz@ozyegin.edu.tr Abstract This study aims to create the very first dependency-to-constituency conversion algorithm optimised for Turkish language. For this purpose, a state-of-the-art morphologic analyser (Yıldız et al., 2019) and a feature-based machine learning model was used. In order to enhance the performance of the conversion algorithm, bootstrap aggregating meta-algorithm was integrated. While creating the conversation algorithm, typological properties of Turkish were carefully considered. A comprehensive and manually annotated UD-style dependency treebank was the input, and constituency trees were the output of the conversion algorithm. A team of linguists manually annotated a set of constituency trees. These manually annotated trees were used as the gold standard to assess the performance of the algorithm. The conversion process yielded more than 8000 constituency trees whose UD-style dependency trees are also available on GitHub. In addition to its contribution to Turkish treebank resources, this study also offers a viable and easy-to-implement conversion algorithm that can be used to generate new constituency treebanks and training data for NLP resources like constituency parsers. Keywords: Constituency parsing, Dependency parsing, Constitueny Dpendency conversion 1. Introduction Following the phrase-based translation model (Koehn et al., 2003), syntactic trees have been prominent in the past studies on statistical grammar (Ding and Palmer, 2005; Ding and Palmer, 2004), statistical parsing (Grammar and Hockenmaier, 2003; Wang and Harper, 2004; Cahill et al., 2008; Candito et al., 2010), machine translation (Huang et al., 2006; Ding and Palmer, 2005; Shen et al., 2008; Xie et al., 2011) and various other subbranches of NLP. There are two main syntactic tree representations: Dependency and constituency trees. Both are able to capture valency, argument structure and various other syntactic relations between phrases. Although it is possible to argue that dependency trees are better equipped to handle long-distance dependen- cies and free word order languages, constituency trees can better illustrate phrasal continuity and government. Based on the notion of dependency grammar first in- troduced by Lucien Tesni`ere (See (Percival, 1990) for a detailed history and discussion), dependency trees con- nect linguistic units (heads and dependents) through di- rected links and labels these links based on the relation- ship between the head and its dependent(s). Inspired by term logic, constituency trees have their theoreti- cal origins at phrase structure grammars (also known as constituency grammars) presented by Noam Chom- sky (Chomsky, 1957). Based on constituency relation, constituency trees illustrate phrases as sub-trees and shows the relationship between not only the head and its dependent but also the relationship between the de- pendents of the same head (See Figure 2). Although …

Content truncated to 500 words